Child care before age two and the development of language

Comments

Transcription

Child care before age two and the development of language
Discussion
Papers
Statistics Norway
Research department
No. 808
May 2015
Nina Drange and Tarjei Havnes
Child care before age two and the
development of language and
numeracy
Evidence from a lottery
•
Discussion Papers No. 808, May 2015
Statistics Norway, Research Department
Nina Drange and Tarjei Havnes
Child care before age two and the development of
language and numeracy
Evidence from a lottery
Abstract
Young children are thought to be vulnerable to separation from the primary caregiver/s. This raises
concern about whether early child care enrollment may harm children's development. We use child
care assignment lotteries to estimate the effect of child care starting age on early cognitive
achievement in Oslo, Norway. Getting a lottery offer lowers starting age by about four months, from a
mean of about 19 months in the control group. Lottery estimates show significant score gains for
children at age seven. Survey evidence and an increase in employment of both mothers and fathers
following the offer, suggest that parental care is the most relevant alternative mode of care. We
document that the assignment lottery generates balance in observable characteristics, supporting
our empirical approach.
Keywords: early child care; child development
JEL classification: J13, J21
Acknowledgements: Thanks to Oslo Municipality for generously providing data, institutional detail
and feedback on the project, in particular to Eli Aspelund, Thomas Bang and Ragnhild Walberg at
HEV. The project received financial support from the Norwegian Research Council (Grant Number
212305 and Grant Number 236947). The project is also part of the research activities at the ESOP
center at the Department of Economics, University of Oslo. ESOP is supported by The Research
Council of Norway (grant no. 179552).
Address: Nina Drange, Statistics Norway, Research Department. E-mail: [email protected]
Tarjei Havnes, Department of Economics, University of Oslo; Statistics Norway, Research
Department, E-mail: [email protected]
Discussion Papers
comprise research papers intended for international journals or books. A preprint of a
Discussion Paper may be longer and more elaborate than a standard journal article, as it
may include intermediate calculations and background material etc.
© Statistics Norway
Abstracts with downloadable Discussion Papers
in PDF are available on the Internet:
http://www.ssb.no/en/forskning/discussion-papers
http://ideas.repec.org/s/ssb/dispap.html
ISSN 1892-753X (electronic)
Sammendrag
Bruken av barnehage blant de minste barna har økt i mange land i løpet av det siste tiåret. I 2010 gikk
om lag 43 % av barna under to år i barnehage i USA, og det tilsvarende gjennomsnittet i OECD-land
var 33 %, opp fra 20 % i 2003. I flere land er bruken av barnehage vesentlig høyere, med en andel
over 50 % i land som Danmark, Korea, Nederland og Norge.
Samtidig er både forskere og politikere bekymret for at tidlig barnehagestart, og dermed tidlig
separasjon fra den primære omsorgspersonen (vanligvis mor), kan forårsake stress og angst hos barnet,
og at dette kan ha potensielt negative effekter for senere utvikling (Bowlby, 1969, Mercer, 2006).
Likevel vet vi fremdeles lite om hvordan barnehagen påvirker utviklingen til de minste barna. Dette er
bekymringsfullt for beslutningstagere fordi offentlige midler brukes til å subsidiere barnehagen, men
også for foreldre som skal bestemme om og når de skal la sine barn starte i barnehagen.
I denne artikkelen undersøker vi effekten av barnehage for barn i alderen 1-2 år på senere kognitiv
utvikling i språk og regning (målt ved syv års alder). Det er av avgjørende betydning at vi forstår
hvilke effekter barnehagen har for denne aldersgruppen, både fordi bruken av barnehage for disse
barna øker sterkt, og fordi barn er antatt å være spesielt sårbare i denne perioden.
Vi utnytter et randomisert lotteri som brukes i Oslo kommune for å fordele tilbud om barnehage når
det er flere barn som søker enn det er tilgjengelige plasser. Dette gir tilfeldig variasjon i alder ved
barnehagestart. Vi dokumenterer at den tilfeldige tildelingen genererer to grupper av barn som ser
svært like ut på observerbare kjennetegn, noe som støtter vår empiriske tilnærming. Vi har tilgang til
unike data på alle som søker plass, hvem som får tilbud og hvem som starter. Dessuten har vi tilgang
til kartleggingsprøver i språk og regning når barna er syv år. Vi bruker den tilfeldige fordelingen av
barnehageplasser som en instrumentvariabel for barnets alder når han eller hun først starter i
barnehagen. Vi benytter data for kohortene 2004-2006, i denne perioden var det langt flere som ønsket
barnehage for sine ettåringer enn som faktisk fikk plass.
Våre resultater viser at barn som tilfeldig får et tilbud om barnehage presterer bedre på både språktesten (0,12 SD) og testen i regning (0,12 SD). Mens barn som «vinner i lotteriet» og får tilbud om
barnehageplass starter i barnehagen ved ca 15 måneders alder, er gjennomsnittsalderen ved barnehagestart for barn som ikke får et tilbud ca 19 måneder. Blant barna som ikke får et tilbud, starter om lag to
tredeler et helt år senere.
Kartleggingsprøvene vi benytter som mål på barnas utvikling er konstruert for å plukke opp barn som
scorer lavt, så et flertall av barna scorer høyt på prøvene. Dette medfører en skjev fordeling av
testresultatene, og vi må derfor vise varsomhet når vi skal fortolke resultatene. Én måte å gjøre dette
på, er å sammenligne våre estimater med gapet mellom testresultater for enkelte veldefinerte
undergrupper av befolkningen. For eksempel tilsvarer den estimerte effekten av å få et tilbud omtrent
det gjennomsnittlige gapet i testprestasjoner mellom gutter og jenter, eller mellom 30 % og 40 % av
gapet mellom barn med lavt og høyt utdannede foreldre. Merk likevel at testresultatene vi benytter kan
betraktes som toppsensurerte, noe som betyr at gapene vi observerer sannsynligvis er mindre enn hva
vi ville ha sett med en mer symmetrisk fordeling av testresultater.
Et viktig argument for å subsidiere barnehage, er at barnehagen kan bidra til å utjevne forskjeller i
kognitiv utvikling mellom barn fra ulike sosioøkonomiske bakgrunn. Vi finner delvis støtte for dette i
våre data: Vi finner sterkere effekter av tidlig barnehage blant barn fra familier med lav inntekt, og
ingen effekt blant familier med høy inntekt, verken i språk eller regning. Dette kan tyde på at subsidier
til barnehage kan være mer effektive dersom de rettes mot husholdninger med lav inntekt.
3
I studier av hvordan barnehagen påvirker barns utvikling, er det viktig å belyse hvilken type barnepass
barna ville hatt dersom de ikke hadde fått barnehageplass. Vanligvis vurderer man alternativene
foreldreomsorg, og andre, mer uformelle, kilder til omsorg. For å undersøke hvilket alternativ som var
mest aktuelt for barna i vår studie, starter vi med å se på resultater fra en spørreundersøkelse fra 2002
der man spurte etter etterspørselen etter og faktisk bruk av barnehage blant foreldre med små barn.
Mens opp mot 70 % av foreldrene oppgir at de etterspør barnehage, enten på heltid eller deltid, er det
bare 33 % som faktisk har sine små barn i barnehagen. Til sammenligning er det 56 % av foreldrene
som sier at de passer på barna sine selv, mens bare 17 % faktisk foretrekker denne løsningen. Dette
tyder på at foreldreomsorg er det dominerende alternativet for norske småbarn generelt.
For å få et anslag på foreldrenes omsorg kontra mer uformell barnepass undersøker vi virkningen av å
få et tilbud om barnehage på foreldrenes arbeidstilbud. Resultatene tyder på at tilbud om barnehage
øker arbeidstilbudet noe blant både mødre og fedre.
4
1
Introduction
Child care enrollment of toddlers has increased in many countries over the last decade.
In 2010, the enrollment of children below age two stood at 43 % in the United States, and
at 33 % in OECD countries, up from 20 % in 2003. In several countries, enrollment is
much higher, with rates above 50 % in countries like Denmark, Korea, the Netherlands and
Norway.
1
At the same time, there is concern among both researchers and policymakers
that separation from the primary caregiver, typically the mother, may cause stress and
anxiety in the child, with potentially adverse eects on children's development [Bowlby,
1969, Mercer, 2006].
3
is largely missing.
2
Yet, evidence on how child care aects the development of toddlers
This is worrying for policymakers, because programs are often heavily
subsidized, but also for parents who need to decide whether and when to enroll their
children in child care.
In this paper, we provide rst evidence on the impact of child care enrollment for children age 12 years old (henceforth toddlers), on their cognitive performance in language
and mathematics at age seven. Determining the impact in this age group is of key importance, both because enrollment rates are increasing at a strong rate and because children
are thought to be particularly vulnerable during this period. For identication, we exploit
a randomized lottery used by the city government in the Norwegian capital Oslo in order
to allocate oers of child care places when child care institutions are over-subscribed, similar to the strategy used by Abdulkadiroglu et al. [2011]. This should provide variation in
child care starting age that is as good as random. Indeed, we document that the allocation mechanism generates balance in observable characteristics, supporting our empirical
approach.
Armed with unique data on all applicants, oers and enrollment, as well as
1 Source:
OECD Family data base.
[1969] denes the attachment phase as the period when the child is from 68 months to 24
months old. The age period we study in this paper is largely around 1224 months.
3 We are aware of only one other study that estimates how child care aects the development of toddlers.
Felfe and Lalive [2014] estimate marginal treatment eects of child care attendance before age three using
county level variation in child care coverage rates in West Germany. They nd positive impacts of child
care on the youngest children, boys, and children from low SES families. A related strand of literature
may indirectly reect the eects of child care attendance in looking at the eect of parental leave policies
on child outcomes (e.g. Dustmann and Schønberg [2012], Liu and Skans [2010], Carneiro et al. [2015]).
The alternative to parental care in most of this studies is, however, likely to be informal and not formal
sources of care.
2 Bowlby
5
performance tests in language and mathematics at age seven, we use the randomized oer
of a public child care slot as an instrumental variable for the age of the child when he
or she rst attends child care. Specically, we consider cohorts born 20042006, among
whom there was substantial oversubscription of toddlers to child care institutions in Oslo.
Our results show that children who randomly get an oer of public child care perform
better on both the language test (0.12 SD) and the mathematics test (0.12 SD) at age 67.
Next, while children that get an oer rst attend child care at about 15 months of age on
average, children who randomly do not get an oer rst attend at about 19 months of age
on average. Among children delayed, about two thirds start a full year later. When we
estimate the impact of child care starting age using the lottery oer as an instrumental
variable, we nd that starting child care one month earlier causes an improvement in test
scores of 0.027 SD and 0.028 SD in language and mathematics, respectively.
Because the distribution of the test score we consider is skewed, we must be careful
when interpreting the economic signicance of our estimates. One way to do this is to
compare our estimates to the gaps in test scores that we observe between well dened
sub-groups of the population.
For instance, the estimated impact of getting an oer
corresponds to about the average gap in test performance between boys and girls, or to
between 30 % and 40 % of the performance gap between children of low and high educated
parents. One should keep in mind, though, that the test scores we consider are essentially
truncated at the top, which means that the observed gaps are likely smaller than what
we would observe with more symmetrically distributed test scores.
One important argument for why governments want to subsidize child care, is that
they can help counter dierences in school readiness between children from dierent socioeconomic backgrounds. When estimating in subgroups, our evidence partly supports
this case: We nd stronger eects of early child care start among children from low income
families, and no impact among high income families in neither language nor mathematics.
This suggests that child care policies may be more eective if targeted at low income
households.
Our results on how child care aects the development of toddlers complement the
6
growing recent literature on how child care institutions aect the development of preschool
age children.
4
The literature is divided in two distinct branches, one focussed on targeted
programs, and another focussed on universal programs available to the general population.
5
While studies of targeted programs often nd positive eects,
6
programs is smaller and ndings are mixed.
the literature on universal
Perhaps as a consequence, the discussion
on child care policies is based largely on the targeted literature and descriptive evidence,
even when the policies considered are universal. In contrast, we study the impact of a
universally available program among applicants that are explicitly not prioritized in child
care.
In studies of how child care aects child development, it is crucial to shed light on
the alternative mode of care that children would have been exposed to if they did not
attend child care. Typically, one considers three alternatives: Parental care, formal child
care and other, more informal, sources of care [Blau and Currie, 2006]. To investigate the
counterfactual mode of care for children who get an oer of a public child care slot, we
start by considering survey data on stated demand and actual use for the population of
parents with toddlers. While about 70 % of parents state demand for child care, either
full time or part time, only 33 % actually have their children enrolled in child care. In
comparison, while 56 % of parents say that they care for their children themselves, only
17 % actually prefer to do so. This suggests that parental care is the dominant alternative
for Norwegian toddlers in general.
To get an estimate of the importance of parental care as the counterfactual in our
particular sample, we next consider the impact of getting an oer of a public child care
4 For
recent reviews, see Almond and Currie [2010], Ruhm and Waldfogel [2011], or Baker [2011].
Perry Preschool and Abecedarian programs are examples of targeted randomized programs (see
Barnett [1995] and Karoly et al. [2005] for surveys of the literature.), while the US Head Start program
provides an example of a targeted non-randomized program (see e.g. Currie [2001] or McKey et al. [1985]
for a review of the ndings). While the picture is somewhat mixed, the most robust evidence on Head
Start tends to show positive eects on long-run outcomes such as high school dropout, college attendance
and crime [Currie and Thomas, 1995, Garces et al., 2000, Ludwig and Miller, 2007, Deming, 2009].
6 Several studies from Canada show a negative impact on a variety of child outcomes [Baker et al.,
2008, Lefebvre and Merrigan, 2008a, DeCicca and Smith, 2013], while Cascio [2009] and Gupta and
Simonsen [2010] nd essentially no impact from child care programs in the US and Denmark, respectively.
In contrast, positive impacts on a number of outcomes are found from child care programs in several
countries, including the US [Fitzpatrick, 2008], Uruguay [Berlinski et al., 2008], Norway [Havnes and
Mogstad, 2011b, 2014], Germany [Dustmann et al., 2013, Felfe and Lalive, 2014], and Spain [Felfe et al.,
2015].
5 The
7
slot on parents employment. Results suggest that the oer increases employment of both
mothers and fathers. While mothers labor supply increases around the margin of full time
employment, fathers labor supply increases mostly at the top.
parents responding to the oer by reducing the
This is consistent with
position of mothers and the work hours of
fathers. This evidence contrasts previous ndings for preschool aged children in Norway,
where eects on parental labor supply have been estimated to be quite weak [Havnes and
Mogstad, 2011a].
7
At the same time, this may not come as a surprise, since parents of
preschool children may both have stronger options in the informal market and be more
willing to use them, compared to parents of toddlers. Also, our estimates relate to the
mid-2000s, while Havnes and Mogstad [2011a] estimate the impact of the introduction
of child care in Norway from the late 1970s.
With much higher employment rates of
Norwegian women in recent years, informal sources of care may well be less available.
The paper proceeds as follows. We rst discuss the institutional background in Section
2. Section 3 presents and discusses our empirical approach, before Section 4 describes our
data. Section 5 presents our main results, including discussions on heterogeneous impacts
and mechanisms, while section 6 concludes.
2
Institutional background
In this section, we provide brief instititutional background about care of young children
in Norway, with a focus on the child care sector.
Parental leave. In 2005, Norwegian parents were entitled to 43 weeks of parental leave
with full wage compensation (alternatively 53 weeks with 80 % compensation).
8
This was
expanded to 44 (54) weeks in 2006. Parents are further entitled to one year each of unpaid
7 Evidence
from other countries is mixed. In a survey of the early literature, Blau and Currie [2006]
report elasticities of maternal employment with respect to the price of child care ranging from 0 to -1.
More recently, Baker et al. [2008] nd a positive eect on maternal labor supply following the introduction
of heavily subsidized universally available child care in Quebec. Meanwhile, Lundin et al. [2008] nd no
such eect when studying a childcare reform which capped childcare prices in Sweden. See also Schlosser
[2005], Cascio [2009], Havnes and Mogstad [2011b], Lefebvre and Merrigan [2008b] and Berlinski and
Galiani [2007]. For a review of the literature, see Blau and Currie [2006].
8 This entitlement is conditional on maternal employment during at least six of the ten months before
the birth. About 85 % of new mothers satisfy this requirement (NOU 2012:15). Remaining parents are
entitled to unpaid parental leave with employment protection, and receive a one-time payment of about
35,000 NOK.
8
leave in immediate continuation of regular parental leave. In practice, most parents can
thus stay at home with their newborn for about a year.
Structure and content of child care in Norway.
To help interpret our results, we
must understand the type of care we are studying.
Child care in Norway is heavily
regulated, with provisions on sta qualications, number of children per adult and per
teacher, size of play area, and educational orientation. Institutions are run by an educated
pre-school teacher responsible for day-to-day management and educational content. The
pre-school teacher education is a one year college degree, including supervised practice in
a formal child care institution. The head teacher is responsible for planning, observation,
collaboration and evaluation of all activities.
The head teacher is also responsible for
communication and collaboration with parents and local authorities, including health
stations and child welfare services when necessary.
Child care regulations specify that
there should be at least one educated pre-school teacher per 10 children aged below three.
In addition, regulations specify that there should be one adult per three children below
three, including the teacher. There is no educational requirements for the additional sta.
In Oslo, about 60 % of child care institutions are public, while the remaining are privately
operated. Both public and private institutions require municipal approval and supervision
to be entitled to federal subsidies that cover around 80 % of costs. Since 2003, parental
copayment is capped at around 2,500 NOK per month for a full time slot.
Child care
institutions are typically open from around 7.30 am to 5 pm.
In terms of educational content, a social pedagogy tradition has dominated child care
practices in Norway since its inception in the 1970s. According to this tradition children
should develop social, language and physical skills mainly through play and informal
learning.
9
The informal learning is typically carried out in the context of day-to-day
social interaction between children and sta, in addition to specic activities for dierent
age groups.
In table I, we report some institutional characteristics of the institutions in our sample.
9 The
social pedagogy tradition to early education has been especially inuential in the Nordic countries and Central-Europe. In contrast, a so-called pre-primary pedagogic approach to early education
has dominated many English and French-speaking countries, favoring formal learning processes to meet
explicit standards for what children should know and be able to do before they start school.
9
We see that an average institution in our sample services 17 children aged 02, with about
three adults per ten children, including one teacher, in line with federal regulations. The
minority share among all children in the institution is about 25 %, reecting the high
share of children with a minority background residing in Oslo. The enrollment of children
with an immigrant background is, however, very low for children below three years of age
[Drange and Telle, 2015]. Note that this implies that our results should be interpreted to
reect the impacts on native children.
Table I: Institutional characteristics
Mean SD
N
Teacher/children
.077
.026
236
No. of adults
15.4
7.6
262
Adult/children
.298
.068
236
Minority share
.256
.207
236
No. of children 02
17.1
10.7
262
Source: The child care register, Statistics Norway.
Child care centers in Norway in an international perspective.
The provision of child
care in Norway bears resemblance with the other Nordic countries with relatively high
public subsidies.
10
However, the enrollment of children below 3 in Norway in 2004 was
44 %, substantially lower than for example Denmark with an enrollment at 83 % . This is
comparable to US enrollment which stood at 38 % for this age group at the time [OECD,
2006].
3
Empirical strategy
We are interested in the eect of early child care enrollment on childrens cognitive
performance, and follow closely the approach in Abdulkadiroglu et al. [2011]. Because the
eects of enrollment are likely to depend on the age at which the child rst attends child
10 For
children below three years old parental contribution in the Nordic countries varied from 915 %
compared to an OECD average of 2530 % in the mid 2000. In 2003, the state subsidy to a child care
slot for a child below three was 9,773 EURO annually in Norway [OECD, 2006].
10
care, we start with the following outcome equation,
yit = γAGEit + X 0it β + it
where
(1)
t denotes the cohort, and AGEit is the age of the child in months when he or she rst
attends any child care institution, public or private.
Xit
are a set of socio-demographic
11
characteristics of the child and parents, measured the year before the child was born.
Because enrollment in child care is likely determined in part by parental preferences
and child innate characteristics, starting age is likely to be correlated with unobserved
determinants of cognitive performance. For instance, we might expect more able parents
to be more closely tied to the labor market, and therefore enroll their children in child
care earlier. If so, then we may expect that children who are enrolled early would perform
better in any case. On the other hand, we might expect more child-centered parents to
enroll their children in child care later. If so, being enrolled early could be a marker for a
poor home environment, which would suggest that these children should perform worse.
This implies that estimation of equation (1) will give biased estimates of the impact of
child care starting age on cognitive performance.
To circumvent this problem, we take advantage of an assignment lottery used by
the Oslo city administration to distribute oers to applicants when institutions were
oversubscribed. Each year, the vast majority of available child care slots in both public
and private institutions are allocated in a centralized allocation round. The application
deadline is around March 1 of each year, for enrollment in mid-August.
Parents may
apply for placement in up to seven child care centers in their application, and may list
both public and private institutions.
Allocation takes place inside the city district of residence, but available slots may be
allocated to children from other city districts after the main allocation round. Children
may be awarded priority placement if they have, for instance, a sibling in the same child
care instititution or are disabled. In our sample, 24 % of children get priority placement.
11 Child
characteristics include gender, month of birth and birth order. Parental characteristics include
dummy variables for full time work, receipt of social assistance, high school completion, college degree,
missing parental education, and missing parent identier.
11
Children that have their rst birthday after September 1 are not included in the main
allocation round, but may receive oers after this round is over. In our analysis we exclude
both of the former groups, to focus on the main group of children that are included in
the main allocation round without being assigned priority.
Based on the applications received, the municipality generates lists of non-priority
applicants to each institution. Lists for private instititutions are transmitted to the institutions, which handle their own admissions based on these lists along with full details
of the individual child and application.
In line with Abdulkadiroglu et al. [2011], we
therefore exclude from our analysis children that have a private institution ranked rst
on their application.
The mechanism for assignment to public institutions resembles a serial dictatorship:
The order of children on the full list of applicants to each public institution is randomized
in the computer before they are presented to the city ocial.
Available slots are then
allocated according to the random rank on the application list, and oers are sent to
parents. Parents may accept or reject the oer. If they reject, the oer is conferred to
the highest ranked child on the application list who did not already get an oer at this
or some other institution. Once a child receives an oer for a child care place, the child
is taken out of the lists to other institutions to which it applied. The child may, however,
maintain their application to the rst ranked institution. This along with the upper limit
on institutions on the list could raise concern about strategic application behavior, but
we nd no suggestions of this below.
The main allocation round ends each year around June 1. After the main allocation
round, available slots may be oered to any applicant, whether or not they ranked the
institution on their application. This process is largely at the discretion of the city ocials
or even child care managers, and is therefore susceptible to manipulation. We therefore
use only oers dated before June 1 each year in our analysis.
To help identify the eect of early child care enrollment, we use an IV strategy, where
we let the oers generated in the assignment lottery act as instruments for child care
starting age. Specically, we specify our IV-model as follows, where equation (2) is the
12
second stage and equation (3) is the rst stage.
yit = γAGEit +
X
αkt Dkit + X 0it β + it
(2)
k
AGEit
=
X
πOF F ERit +
ηkt Dkit + X 0it ψ + ωit .
(3)
k
where
OF F ERit
is a dummy equal to one if the child received an oer of a public child
care place generated in the assignment lottery. In both the rst and second stage, residuals
are clustered at the level of the rst choice institution. Following Abdulkadiroglu et al.
[2011], we also include indicators for lottery-specic risk sets
Dkit
to account for the
fact that children apply to dierent institutions with dierent numbers of applicants and
available slots. The extent of oversubscription determines the probability of receiving a
lottery oer.
If oversubscription rates are correlated with, for instance, the quality of
the child care institution, and applying to good institutions in turn is correlated with
unobservable traits that determine cognitive performance, then a comparison based on
lottery oers may give biased estimates of the impact on cognitive performance of early
child care enrollment. To guard against such bias, we control for the number and identity
12
of institutions to which an applicant applied.
In order for random oers to be relevant, we need to have over-subscription of toddlers
to child care institutions. This is determined by the number of non-priority applicants
per remaining available slot after priority placements. Table II shows descriptive statistics
for the number of available slots, the number of applicants and oversubscription to child
care institutions in Oslo in the period we consider. Oversubscription is both strong and
widespread:
The mean number of applicants to each child care slot is just under 12,
while the median is nine. This is mirrored in the fact that only 27 % of the children in
our sample get an oer in the assignment lottery, and in the strong rst stage estimates
documented below.
The validity of the lottery oer as an instrument for child care starting age, relies on
12 Specically,
the risk set includes a full set of dummy variables for each institution by year, so that for
each institution and year there is a dummy equal to one if child i applied to that institution in that year
and zero otherwise. In addition, the risk set includes dummy variables for the number of applications by
year.
13
Table II: Applications, places and oversubscription in child care institutions in the centralized admission process in Oslo, 20052007.
Mean
No. of places
SD
Min Max
6.21
4.75
0
33
No. of applicants
50.75
30.49
1
165
Applicants Places
44.55
29.47
1
158
Applicants / Places
11.91
11.56
1.5
109
the quality of the assignment lottery. While the city administration ensures us that the
lottery was randomized by a computer algorithm, as described earlier, there is always
the possibility that the randomization failed, or that there was manipulation between the
actual randomization and the sending out of oers. To verify that the randomization was
successful, the rst four columns of Table III reports means and standard deviations of
background characteristics for children in our estimation sample, separately by whether
the child received an oer or not. Table III shows that the two groups look well balanced.
We also test this formally in the context of our econometric model, by regressing the oer
dummy on all characteristics, controlling for the risk sets. The nal column of Table III
reports
t-statistics
of the individual coecients from this regression, which are usually
very low. In a joint test of whether coecients on all covariates are equal to zero, we get
an
F -value
of
1.22,
conrming that the two groups are indeed well balanced.
14
Table III: Balance in background characteristics between children with and without a
lottery oer.
No oer
Oer
t
-value
Girl
0.497
0.500
0.523
0.500
0.11
Age
14.115
1.949
14.163
2.017
1.06
0.10
0.300
0.096
0.295
-1.34
14.793
3.060
14.731
2.923
0.69
Immigrant
Mother
years of educ.
earnings
297 323
162 341
292 643
159 379
0.67
age
33.559
4.3749
33.386
4.439
-2.09
age rst birth
29.900
4.5291
29.860
4.580
0.77
Father
years of educ.
14.532
3.460
14.417
3.554
-0.07
428 276
388 353
427 315
447 764
0.71
age
35.512
6.337
35.128
7.328
1.71
age rst birth
31.552
5.880
30.822
6.474
-0.60
earnings
N = 1, 425
N = 585
F = 1.22
Note: The table reports means and standard deviations of covariates by whether the child received an
oer in the assignment lottery. The nal column reports t-statistics of the individual coecients from
a regression of the oer dummy on all characteristics, controlling for the risk sets, and the F -statistic
from a joint test of whether coecients on all covariates are equal to zero. Age refers to the age in
months of the child in August of the year of application. Earnings are pensionable income from work
and self-employment. Detailed descriptions of the background characteristics are provided in 4.
4
Data
Data.
Our data are based on several dierent administrative registers from the Oslo
city government and Statistics Norway. Firstly, we have access to the municipal database
used in the centralized application system for child care in Oslo. This provides information
on applications for and enrollments in virtually all child care institutions in Oslo for the
years 20052010, including both public and private child care institutions.
Applicants
that list several institutions in their applications are registered as separate coincident
applications. The database also provides information about oers of slots in public child
care centers. Applications, enrollment and oers are recorded with date of receipt, date
of rst attendance and date the oer was made, respectively.
Second, we have access to a database with information about performance on tests
made available by the school authority in Oslo municipality. This provides information
15
about enrollment in primary school and score on performance tests in Norwegian language
and mathematics, conducted in April of rst grade.
The tests are designed nationally,
and are intended to help identify underperforming children, enabling schools to allocate
resources to these children. The language test maps the ability to write letters, recognize
written letters, identify spoken letters, combine sounds, write words, read words and read
sentences. The mathematics test maps the ability to count, to compare numbers, to rank
numbers, to recognize sequences of numbers, to count forward and backward from a given
number, to split a number into two other numbers (i.e.
assignments and to add two numbers.
4 = 1 + . . .),
to solve textual
We provide further detail on these tests in the
appendix.
Each test is scored on a relatively ne scale, where students may score from zero to
105 in language, and zero to 50 in mathematics. Because tests are designed to identify
children with problems, test score distributions are skewed,
13
with about ten and 15 % of
children in our sample getting the top score in language and mathematics, respectively.
This is important to keep in mind when interpreting our results.
In our analysis, we consider two outcomes from each test.
First, we normalize the
scores to have mean zero and standard deviation equal to one. Second, we use dummy
variables for performance below a nationally determined threshold. Thresholds are set for
individual parts of each test from a trial of the test on a panel of children, conducted prior
to actual testing each year. The thresholds are intended to identify the bottom 1520 %
of children. From these we dene the dummy variables
Below threshold
equal to one if
the child has one or more test parts with scores below the threshold, and zero otherwise,
separately for language and mathematics.
Finally, as a summary measure of cognitive
skills, we also consider the unweighted average of the standardized test scores in language
and mathematics.
Third, we can link both databases to rich Norwegian administrative registers available from Statistics Norway, with individual information on demographics (e.g. sex, age,
immigrant status, marital status, number of children), socioeconomic status (e.g. years
13 Appendix
gure (A1) draws the distribution of test scores in our sample.
16
of education, income, employment status), and residence. Income and employment data
are collected from tax records and other administrative registers. The household information is from the Central Population Register, which is updated annually by the local
population registries and veried by the Norwegian Tax Authority. We also have access
to national registry data on municipal child care coverage reported by the child care institutions themselves. The reliability of Norwegian register data is considered to be very
good, as is documented by the fact that they received the highest rating in a data quality assessment prepared for the OECD by Atkinson et al. [1995]. Importantly, all data
sources contain personal identiers that allow us to link individuals across all registers.
Estimation sample. We start with the universe of children born 20042006,
14
for whom
parents apply for a child care slot in Oslo for the rst time the calender year they turn one
year old. Because our identication comes from oers of public child care slots, we focus
attention on children with a public institution on the rst rank, while we allow both private
and public institutions on ranks 27. As discussed above, we also exclude children who had
priority in child care or who turn one after September 1 in the application year, since our
identication does not inuence these children. We nally exclude a handful of children
with missing values on our dependent variables, and a handful of children registered as
starting in child care before ten months old. Rather than excluding children with missing
values on control variables, we construct dummy variables for missing and include these
in our regressions. Our nal estimation sample then consists of 2,010 children.
5
Empirical results
We now turn to our main analysis of how early enrollment in child care aected the
cognitive performance of children at age seven.
We start by a reduced form analysis,
where we compare outcomes of children who got a lottery oer to children who did not
get a lottery oer. Table IV shows means and standard deviations of our main outcome
variables and our key explanatory variable, child care starting age. The mean test scores
14 Due
to a restrictive storage policy in the municipality, data on children born in January and February
2004 were deleted from the application data base before we got access to it. We are therefore not able to
include these children in our sample.
17
Table IV: Performance in language and mathematics tests at age seven for children with
and without a lottery oer.
No oer
Oer
Average score
71.84
(8.10)
72.49
(7.00)
Language
98.90
(11.26)
99.75
(8.95)
below limit
0.132
(0.339)
0.118
(0.323)
Mathematics
44.78
(6.55)
45.24
(6.31)
below limit
0.067
(0.251)
0.055
(0.228)
Starting age (months)
18.88
(8.06)
15.03
(4.59)
N = 1, 425
N = 585
Note: Oer are children that received an oer in the assignment lottery, while No oer are children that
did not receive an oer in the assignment lottery, see Section 3.
show that children who receive a lottery oer perform about eight to nine percent of a
standard deviation better than children who do not receive a lottery oer. Meanwhile,
just over 12 % of children are below the threshold for low performance in language, while
about six percent are below the threshold in mathematics. In both subjects, the mean
child who got a lottery oer performs about half a point better than the mean child who
did not get a lottery oer, and are about 1.5 percentage point less likely to score below
the threshold for low performance. This is rst evidence that early child care enrollment
has a positive impact on childrens cognitive development.
Next, we consider this reduced form model formally, by estimating the impact of
getting an oer on test performance, controlling for risk sets as in equation (3).
report estimates both including and excluding covariates.
We
While including covariates
should not change our estimates when the explanatory variable of interest is as good as
random, it may be helpful to improve precision in our estimates. Results reported in Table
V clarify our observations from above, and indicate that getting a lottery oer improved
the average performance of children by about 12 % of a standard deviation. This eect
was driven both by an improvement in language of about 12 % of a standard deviation,
and by an improvement in mathematics of about the same magnitude.
Table VI reports estimates from our full IV-model, where the receipt of a lottery oer
is used to instrument for the age at rst attendance in a child care institution. The two
last rows of the table report estimates from our rst stage equation, and show that the
lottery oer decreased starting age by about four months on average.
18
The
F -statistic
Table V: Estimates of the impact of a lottery oer on performance in language and
mathematics.
No controls
b
SE
Average score
0.121
Language
0.118
below limit
Mathematics
below limit
With controls Mean
b
SE
(0.064)
0.105
(0.061)
0.000
(0.076)
0.104
(0.074)
0.000
-0.025
(0.024)
-0.028
(0.023)
0.128
0.124
(0.066)
0.106
(0.062)
0.000
-0.023
(0.014)
-0.020
(0.014)
0.064
Note: Eects are reported as percent of the standard deviation. Standard errors are heteroskedasticity
robust and clustered at the rst choice institution level. All regressions include a risk set with a full set
of dummy variables for each institution by year and the number of child care institutions listed. We also
include cohort xed eects. Column 2 reports estimates without covariates whereas column 4 reports
estimates including the controls listed in Table III.
on the instrument is about 100, which implies that we need not worry about problems
associated with weak instruments. To understand more in detail how getting a lottery
oer aects child care starting age, Figure I shows the cumulative distribution of children
having started child care at dierent ages. While 91 % of children who received an oer
had started child care by 18 months old, this was the case for only 65 % of the comparison
group. Among those children who were delayed, more than two thirds started a full year
later; 96 % (99 %) of children who received an oer had started child care by 24 (36)
months old, compared to 77 % (95 %) in the comparison group.
Turning now to the IV-estimates in Table VI, the estimates without controls suggests
that starting child care one month later causes a drop in school performance of just under
three percent of a standard deviation. This is driven by a drop in both the language score
and the mathematics score of just under 0.03 SD. All of these estimates are signicant at
the ve percent level. When we consider the impact on the probability of scoring below
the limit for low performance, we nd an impact of around 0.6 percentage points on both
tests.
As expected from the above balancing analysis, estimates barely move when we
include covariates.
To understand the economic signicance of changes in test scores is in general somewhat dicult, because they do not have a meaningful cardinal scale [Cunha and Heckman,
2008]. This may be particularly true in our case, where the distribution of the test score
19
1
.8
.6
.4
.2
12
24
36
48
Starting age (months)
Offer
No offer
Figure I: Cumulative distribution of age at child care start for children with and without
a lottery oer.
is skewed, and quite dierent from the often bell-shaped test scores considered in the
literature.
Though comparisons of estimates across dierent test score outcomes is al-
ways risky, this means that it could be particularly misleading to compare our estimates
directly to those found in other studies.
To interpret our estimates, we need to map them into a metric that is more easily
interpretable in other contexts. One way to do this is to compare our estimates to the
gaps in test scores that we observe between well dened sub-groups of the population.
For instance, the average gender gap in performance between boys and girls in our sample
is about 0.12 SD on both the language and the mathematics test (cf. table VII below).
Our estimates suggest, therefore, an improvement comparable to the gender performance
gap when a child gets a lottery oer. In mathematics, winning the lottery is predicted to
improve performance by about 0.12 SD, which correspondsto between 30 and 40 % of the
performance gap between children of low and high educated parents. Remember, though,
that the test scores we consider are essentially truncated at the top here, so the observed
gaps are likely smaller than what we would observe with more symmetrically distributed
test scores.
20
Table VI: IV-estimates of the impact of child care starting age on performance in language
and mathematics.
No controls
b
SE
Average score
-0.027
Language
With controls Mean
b
SE
(0.011)
-0.025
(0.011)
0.000
-0.027
(0.013)
-0.025
(0.013)
0.000
below limit
0.006
(0.004)
0.007
(0.004)
0.128
Mathematics
-0.028
(0.011)
-0.025
(0.011)
0.000
below limit
0.005
(0.003)
0.005
(0.003)
0.064
Oer
-4.430
(0.443)
-4.224
(0.428)
17.760
First stage
F-value (instrument)
100.0
97.4
Note: Eects are reported as percent of the standard deviation. Standard errors are clustered at the
rst choice institution level and robust to heteroskedasticity. All regressions control for the risk set
by including a full set of dummy variables for each institution by year and the number of child care
institutions listed in the application, see Section 3. Control variables are listed in Section 3.
5.1 Heterogeneous eects of early child care enrollment
One important argument for why governments want to subsidize child care, is that
they can help counter dierences in school readiness between children from dierent socioeconomic backgrounds. It is therefore natural to next consider whether the starting
age in child care has a dierent impact on toddlers from dierent socioeconomic groups.
Table VII reports estimates from our IV-model including covariates, for children in
dierent groups. Most strikingly, we nd stronger eects among children from low income
families.
Indeed, among high income families, we nd no impact of early child care
start on performance in neither language nor mathematics. This suggests that child care
policies may be more eective if targeted at low income households. Across other groups,
estimates are broadly similar. The exception is parental education, where children from
high educated parents are estimated to improve in mathematics from early child care
enrollment, while there is no impact among children from low educated parents.
5.2 Mechanisms
The eect of child care enrollment is related to the alternative mode of care had the
children not been enrolled in child care.
In our case, our instrument pushes child care
21
Table VII: IV-estimates of the impact of child care starting age on test performance in
dierent subgroups.
b
Language
SE
Mean
b
Mathematics
SE
Mean
N
Boys
-0.029
(0.018)
-0.066
-0.064
(0.019)
0.058
996
Girls
-0.025
(0.028)
0.065
-0.057
(0.025)
-0.057
1,014
Parents educ. low
-0.018
(0.032)
-0.136
0.001
(0.025)
-0.170
1,043
Parents educ. high
-0.015
(0.009)
0.147
-0.015
(0.014)
0.184
967
Family income low
-0.019
(0.029)
-0.077
-0.026
(0.023)
-0.113
1,005
Family income high
-0.004
(0.013)
0.077
-0.004
(0.016)
0.113
1,005
Mom age low
-0.038
(0.027)
-0.061
-0.033
(0.025)
-0.036
1,087
Mom age high
-0.010
(0.014)
0.072
-0.033
(0.018)
0.042
923
Note: Eects are reported as percent of the standard deviation. Standard errors are clustered at the
rst choice institution level and robust to heteroskedasticity. All regressions control for the risk set
by including a full set of dummy variables for each institution by year and the number of child care
institutions listed in the application, see Section 3. Control variables are listed in Section 3.
enrollment forward by four months on average, or by about one year for a third of the
control group. After this period, children who do not get a lottery oer are also on average
enrolled in child care. For both, enrollment in child care is largely an absorbing state,
with the vast majority of children who enroll staying enrolled until the school starting
age (which is six years in Norway). To understand the drivers behind our estimates, we
therefore need to consider these two modes of care. That is, what type of care are control
children in before they start regular child care, and what are the characteristics of the
formal care that they attend once they do start regular child care.
We start by considering the alternative mode of care that children rst attend if they
do not get an oer. Typically, one considers three alternatives: Parental care, formal child
care and other, more informal, sources of care [Blau and Currie, 2006]. To investigate the
alternative mode of care for children who get an oer of a public child care slot, we start
by considering survey data on stated demand and actual use for the population of parents
with toddlers, presented in Table VIII. While about 70 % of parents state demand for
child care, either full time or part time in combination with other forms of care, only 27 %
actually have their children enrolled in child care. In comparison, while 56 % of parents
say that they care for their children themselves, only 17 % actually prefer to do so. This
22
suggests that parental care is the dominant alternative for Norwegian toddlers in general.
To get an estimate of the importance of parental care as the counterfactual in our
particular sample, we now consider the impact of getting an oer of a public child care
slot on parents earnings and labor participation. Unfortunately, we do not have data on
hours of work. To measure labor market attachment, we therefore rely on information
about annual earnings, including wages and income from self-employment. Specically, we
construct dummy variables for employment based on the basic amounts in the Norwegian
Social Insurance Scheme (used to dene labor market status, determining eligibility for
unemployment benets as well as disability and old age pension).
In 2006, one basic
amount was about 80,000 NOK, or about 13,000 USD. Following Havnes and Mogstad
[2011a], parents are dened as employed if they earn more than two basic amounts and
full time equivalent if they earn more than four basic amounts. Because the child care
year starts in August, the impact on parental labor supply may materialize both in the
fall of the application year, and in the spring of the following year. We have therefore
estimated the impact on outcomes in both years.
Estimates from the reduced form are reported in Table IX. They suggest that receiving
a lottery oer increases labor supply of both mothers and fathers. While mothers labor
supply increases around the margin of full time employment, fathers labor supply increases
mostly at the top. This is consistent with parents responding to the oer by reducing
on
the extensive margin for mothers and on the intensive margin for fathers, as suggested by
anecdotal evidence. This evidence contrasts previous ndings for preschool aged children
in Norway, where eects on parental labor supply have been estimated to be quite weak
[Havnes and Mogstad, 2011a]. This may come as no surprise, since parents of preschool
Table VIII: Survey evidence on the demand for and use of child care.
Stated demand Actual use
Parents
0.17
0.56
Relatives
0.04
0.04
Unlicensed care givers
0.08
0.13
Child care
0.42
0.22
Combined/Other
0.28
0.05
Source: Pedersen [2003].
23
Table IX: Estimates of the impact of a lottery oer on parental labor supply the same
year and the next year.
Mother
b
Application year
SE
Mean
b
Following year
SE
Mean
Earnings
12 192
(14 028)
325 032
26 993
(20 764)
375 982
Employment
-0.012
(0.026)
0.902
0.006
(0.024)
0.910
Full-time eq.
0.042
(0.034)
0.556
0.061
(0.034)
0.680
Earnings
26 986
(54 388)
634 239
100 367
(64 988)
676 801
Employment
-0.0196
(0.018)
0.963
-0.005
(0.0199)
0.969
Full-time eq.
-0.024
(0.027)
0.860
0.0095
(0.0250)
0.881
Father
Note: Eects on earnings are reported in NOK, 1 USD ≈ 6 NOK. Standard errors are clustered at
the rst choice institution level and robust to heteroskedasticity. All regressions control for the risk set
by including a full set of dummy variables for each institution by year and the number of child care
institutions listed in the application, see Section 3. Control variables are listed in Section 3.
children may both have stronger options in the informal market and be more willing to
use them, compared to parents of toddlers. Also, our estimates relate to the mid-2000s,
while Havnes and Mogstad [2011a] estimate the impact of the introduction of child care
in Norway from the late 1970s. With much higher employment rates of Norwegian women
in recent years, informal sources of care may well be less available.
Next, we investigate whether the characteristics of the child care institution that
the child rst attends depend on whether the child received a lottery oer or not.
If
children with oer not only started earlier but also attended better quality child care
institutions, then this could be driving the improvement in performance we observed in
our main estimates. To evaluate this, Table X lists a wide set of characteristics of the
child care institutions that children rst attend. Structural characteristics are mean test
scores of all children who rst attended the institution, the childteacher ratio and travel
distance from home.
Sta characteristics are mean characteristics of sta, while peer
characteristics are mean background characteristics of children in the same institution.
In the rst four columns of Table X, we report the means and standard deviations, while
the nal two columns report reduced form estimates of the impact of a lottery oer on each
characteristic.There are few indications that children with oers attended higher quality
institutions. On the contrary, children with lottery oers seem to attend institutions with
24
Table X: Characteristics of the rst child care institution attended for children with and
without a lottery oer
No oer
Oer
Red. form
b
SE
Mean
SD
Mean
SD
Language
98.668
6.077
98.645
5.358
-0.493
0.357
below limit
44.427
3.766
44.772
3.139
0.128
0.230
Mathematics
0.130
0.169
0.123
0.149
0.008
0.010
below limit
0.083
0.141
0.067
0.102
-0.013
0.010
14.732
7.968
16.950
6.102
1.309
0.675
Distance (km)
3.206
4.031
2.315
3.559
-0.535
0.293
Distance (min)
6.215
7.239
4.330
5.864
-1.066
0.543
Structural characteristics
Children/teacher
Sta characteristics
Income
249,267
41,712
261,939
26,632
13,163
3,379
College graduates
0.361
0.155
0.3570
0.100
0.001
0.012
Immigrants
0.233
0.193
0.2247
0.153
-0.006
0.014
0.110
0.115
0.1053
0.100
-0.006
0.010
36.913
6.354
39.2731
5.137
1.699
0.524
Males
Age
Peer characteristics
Family income
760,797
273,276
734,690
239,658
-23,174
16,747
College graduates
0.505
0.242
0.474
0.216
-0.003
0.014
Immigrants
0.074
0.117
0.088
0.124
0.010
0.007
Males
0.500
0.196
0.503
0.193
0.009
0.014
Young
0.498
0.217
0.466
0.188
-0.023
0.014
Note: Oer are children that received an oer from the assignment lottery, while No oer are children
that did not receive an oer from the assignment lottery, see Section 3. Mean reects average characteristics of the child care center that the child rst attends. The rst four columns report means and
standard deviations. The nal two columns report reduced form-estimates from our IV-model, including
all covariates and a full set of dummy variables for each institution by year and the number of child care
institutions listed in the application, see Section 3.
slightly less teachers and adults per child, higher minority shares and larger numbers of
children. At the same time the sta and peer composition is largely similar across the
two groups.
25
45.6
100
No offer
Offer
No offer
99
44.8
99.2
45
99.4
45.2
99.6
45.4
99.8
Offer
12
18
24
30
Starting child care before age (months)
36
(a) Language
12
18
24
30
Starting child care before age (months)
(b) Mathematics
Figure II: Performance in language and mathematics by child care starting age, for children with and without a lottery oer.
Note: Figures show local linear regression estimates of mean language and mathematics performance
against starting age in child care, using an epanichnikov kernel with bandwidth set to one month.
Finally, to evaluate the plausibility of starting age as a mechanism, we consider childrens performance as a function of the actual starting age separately among children who
received a lottery oer and among children who did not receive an oer. In Figure I above
we saw the cumulative distribution of children having started child care at dierent ages.
Figure II reports the mean test score among children who started child care before different ages, approximated by a local linear regression. At the very right of these graphs,
the gap between the two groups corresponds to the dierence in performance among all
treated and all control children, similar to our reduced form eect. As we move left, the
treatment and comparison groups are becoming more homogenous in terms of the age at
which they start child care.
If starting age is an important mechanism, we expect the
performance of children to become more similar as starting ages move closer together.
This is largely conrmed in both panels: Children who start child care early, whether
they get a lottery oer or not, perform similarly on the tests, while children who start
later and that did not get an oer, tend to perform worse. This is striking in the case
of the mathematics test, where the dierence between the groups is entirely driven by
children in the control group that start late. For the language test, the gap also widens
with starting age, but here the gap between early starters suggests that the treatment
26
36
also generates other relevant dierences.
6
Concluding remarks
Child care enrollment of young children is substantial and growing, and child care is
often heavily subsidized by the government. At the same time, there is concern among
both researchers and policymakers that separation from the primary caregiver, typically
the mother, may cause stress and anxiety in the child, with potentially adverse eects
on children's development [Bowlby, 1969, Mercer, 2006]. Yet, evidence on how child care
aects the development of toddlers is largely missing.
In this paper, we present evidence on the impact of early child care enrollment on
the cognitive performance of children at age seven. Results indicate that early child care
enrollment has a benecial eect for children's performance both on a language test and
on a mathematics test. Looking across subgroups, we nd stronger eects of starting child
care early among children from low income families, and no impact among high income
families in neither language nor mathematics. This suggests that child care policies may
be more eective if targeted at low income households.
Our results on how child care aects the development of toddlers extends the growing
recent literature on how child care institutions aect the development of preschool age
children. While results are mixed, several studies have shown positive eects, in particular
for children from disadvantaged families. Our study shows that positive eects of child
care are not unique to preschool children, but can be extended also to toddlers below
18 months of age.
Importantly, our estimates lend no support to the concerns about
detrimental impact of child care at early ages. This is true even though children who are
stopped from starting child care early are likely to be at home with a parent.
References
Atila
Abdulkadiroglu,
Kane,
schools:
and
Parag
Evidence
Joshua
A.
from
D.
Pathak.
boston's
Angrist,
Susan
Accountability
charters
27
and
M.
Dynarski,
and
pilots.
exibility
Thomas
in
J.
public
The Quarterly Jour-
nal of Economics,
126(2):699748,
2011.
doi:
10.1093/qje/qjr017.
URL
http://qje.oxfordjournals.org/content/126/2/699.abstract.
Douglas Almond and Janet Currie. Human capital development before age ve. NBER
Working Papers 15827, National Bureau of Economic Research, Inc, March 2010.
Income distribution in OECD
countries : evidence from the Luxembourg Income Study. OECD Publications and
A. B. Atkinson, Lee Rainwater, and Timothy M. Smeeding.
Information Center, Paris, 1995. ISBN 926414577.
Michael Baker. Innis lecture: Universal early childhood interventions: what is the evidence base?
Canadian Journal of Economics, 44(4):10691105, November 2011.
URL
http://ideas.repec.org/a/cje/issued/v44y2011i4p1069-1105.html.
Michael Baker, Jonathan Gruber, and Kevin Milligan. Universal child care, maternal labor
supply, and family well-being.
The Journal of Political Economy, 116(4):pp. 709745,
2008. ISSN 00223808.
W. Steven Barnett. Long-term eects of early childhood programs on cognitive and school
outcomes.
Future of Children, pages 2250, 1995.
Samuel Berlinski and Sebastian Galiani. The eect of a large expansion of pre-primary
school facilities on preschool attendance and maternal employment.
Labour Economics,
14(3):665680, June 2007.
Samuel
ing
les.
Berlinski,
children
a
Sebastian
better
start:
Galiani,
and
Preschool
Journal of Public Economics,
Marco
Manacorda.
attendance
92(5-6):14161440,
and
school-age
June
2008.
GivproURL
http://ideas.repec.org/a/eee/pubeco/v92y2008i5-6p1416-1440.html.
Pre-School, Day Care, and After-School Care: Who's
Minding the Kids?, volume 2 of Handbook of the Economics of Education, chapter 20,
David Blau and Janet Currie.
pages 11631278. Elsevier, 2006.
John Bowlby.
Attachment and Loss (Vol 1) Attachment. Hogarth, London, 1969.
28
Pedro Manuel Carneiro, Katrine V. Løken, and Kjell G. Salvanes. A ying start? maternity leave benets and long run outcomes of children.
Journal of Political Economy,
Forthcoming, April 2015.
Elizabeth U. Cascio.
Do investments in universal early education pay o ?
long-term
eects of introducing kindergartens into public schools. Working Paper 14951, National
Bureau of Economic Research, May 2009.
Flavio Cunha and James J. Heckman. Formulating, identifying and estimating the technology of cognitive and noncognitive skill formation.
Journal of Human Resources, 43
(4):738782, 2008.
Janet Currie. Early childhood education programs.
Journal of Economic Perspectives,
15:213238, 2001.
Janet Currie and Duncan Thomas. Does head start make a dierence?
American Eco-
nomic Review, 85(3):341364, June 1995.
Phillip DeCicca and Justin Smith. The long-run impacts of early childhood education:
Evidence from a failed policy experiment.
Economics of Education Review, 36:4159,
2013.
David Deming. Early childhood intervention and life-cycle skill development: Evidence
from head start.
American Economic Journal: Applied Economics, 1(3):111134, 2009.
Nina Drange and Kjetil Telle. Promoting integration of immigrants: Eects of free child
care on child enrollment and parental employment.
Labour Economics,
Fortcoming,
2015.
Christian
erage
Dustmann
and
and
children's
plied Economics,
Uta
Schønberg.
long-term
4(3):190224,
outcomes.
2012.
Expansions
in
maternity
cov-
American Economic Journal: Apdoi:
10.1257/app.4.3.190.
http://www.aeaweb.org/articles.php?doi=10.1257/app.4.3.190.
29
leave
URL
Christian Dustmann, Anna Raute, and Uta Schønberg. Does universal childcare matter?
evidence from a large expansion in pre-school education.
Working paper, University
College of London, 2013.
Christina Felfe and Rafael Lalive. Does early child care help or hurt children's development? IZA Discussion Papers 8484, IZA, 2014.
Christina Felfe, Natalia Nollenberger, and Nuria Rodriguez-Planas. Can't buy mommy's
love? universal childcare and children's long-term cognitive development.
Journal of
Population Economics, 28(2):393422, 2015.
Maria D. Fitzpatrick. Starting school at four: The eect of universal pre-kindergarten on
children's academic achievement.
The B.E. Journal of Economic Analysis & Policy, 8
(1), 2008.
Eliana Garces, Duncan Thomas, and Janet Currie.
Longer term eects of head start.
Working Paper 8054, National Bureau of Economic Research, December 2000.
Nabanita
and
30
Datta
universal
43,
Gupta
high
2010.
and
quality
ISSN
Marianne
child
Simonsen.
care.
0047-2727.
doi:
Non-cognitive
child
outcomes
Journal of Public Economics,
94(1-2):
10.1016/j.jpubeco.2009.10.001.
URL
http://www.sciencedirect.com/science/article/pii/S0047272709001169.
Tarjei Havnes and Magne Mogstad. Money for nothing? universal child care and maternal
employment.
Journal of Public Economics, 95(11-12):14551465, 2011a.
Tarjei Havnes and Magne Mogstad. No child left behind. subsidized child care and children's long-run outcomes.
American Economic Journal: Economic Policy, 2011b.
Tarjei Havnes and Magne Mogstad.
Is universal child care leveling the playing eld?
Journal of Public Economics, In press, 2014.
Lynn A. Karoly, M. Rebecca Kilburn, and Jill S. Cannon.
Early Childhood Interventions:
Proven Results, Future Promise. RAND Corporation, Santa Monica, CA, 2005.
30
Pierre Lefebvre and Philip Merrigan. Family background, family income, cognitive tests
scores, behavioural scales and their relationship with post-secondary education participation: Evidence from the nlscy. Cahiers de recherche 0830, CIRPEE, 2008a. URL
http://ideas.repec.org/p/lvl/lacicr/0830.html.
Pierre
of
Lefebvre
mothers
and
with
Philip
young
Merrigan.
children:
nal of Labor Economics,
A
Child-care
natural
26(3):519548,
policy
and
experiment
2008b.
doi:
the
from
labor
canada.
10.1086/587760.
supply
JourURL
http://www.journals.uchicago.edu/doi/abs/10.1086/587760.
Qian Liu and Oskar Nordstrom Skans. The duration of paid parental leave and children's
scholastic performance.
January 2010. URL
The B.E. Journal of Economic Analysis & Policy, 10(1):135,
http://ideas.repec.org/a/bpj/bejeap/v10y2010i1n3.html.
Jens Ludwig and Douglas L. Miller.
Does head start improve children's life chances?
evidence from a regression discontinuity design.
The Quarterly Journal of Economics,
122(1):159208, 2007. doi: 10.1162/qjec.122.1.159.
Daniela Lundin, Eva Mork, and Bjorn Ockert.
push female labour supply?
How far can reduced childcare prices
Labour Economics,
15(4):647659, August 2008.
URL
http://ideas.repec.org/a/eee/labeco/v15y2008i4p647-659.html.
Ruth Hubbell McKey, Larry Condelli, Harriet Ganson, Barbara J. Barrett, Catherine
McConkey, and Margaret Plantz. The impact of head start on children, families and
communities. nal report of the head start evaluation, synthesis and utilization project.
Technical report, U.S. Department of Health and Human Services, 1985.
Jean Mercer.
Understanding Attachment: Parenting, Child Care, and Emotional Devel-
opment. Praeger, 2006.
OECD. Starting strong ii: Early childhood education and care. Technical report, 2006.
Department of Education.
Kartlegging av tallforståelse og regneferdighet på 1. trinn.
laererveiledning. 2011a.
31
Department of Education. Obligatorisk kartleggingsprøve: Kartlegging av leseferdighet
1. årstrinn del 1: Laererveiledning. 2011b.
Silje Vatne Pedersen.
Barnefamiliers tilsynsordninger, yrkesdeltakelse og bruk av kon-
tantstøtte våren 2002. Reports 3, Statistics Norway, 2003.
Christopher J. Ruhm and Jane Waldfogel. Long-Term Eects of Early Childhood Care
and Education.
SSRN eLibrary, 2011.
Analia Schlosser. Public preschool and the labor supply of arab mothers: Evidence from
a natural experiment. Technical report, Mimeo, The Hebrew University of Jerusalem,
2005.
32
A
Additional material
Density
.1
0
0
.05
.05
Density
.1
.15
.15
.2
.2
A.1 Tables and gures
0
20
40
60
80
100
0
10
20
norsk1
30
40
matte1
(a) Language
(b) Mathematics
Figure A1: Distribution of test scores.
A.2 Background information about the language test
Every teacher who is responsible for carrying out tests in his/her class receives a
teacher's instruction manual.
The following text about the test is from this manual
of Education [2011b].
Conditions for learning
Test part 1:
The student's attitude when it comes to reading: This part should provide
information about the student's attitude and interests related to dierent activities related
to the written language. Results on this part of the test should not been given a score
but is meant as information to the teacher.
Level of knowledge about the alphabet
Test part 2:
To write letters: This measures the student's ability to link sound and
letter as well as their ability to construct the letters in question. Firstly, the students hear
a word. Secondly the teacher repeats the rst sound of the word and asks the students
33
50
to write the letter that goes with that sound. There is a picture supporting the word in
the student book. The students' results on this test should be scored by the teacher.
Test part 3:
To recognize letters: This part measures one of the basic skills in reading.
Students have four minutes to their disposal.
With the starting point a capital letter
(versal) the students shall recognize the same lower case letter among several other lower
case letters (minuskler). The students' results on this test should be scored by the teacher.
This test part does not have a critical threshold since many students may have been
exposed to only one type of letter throughout the rst school year. The score registrations
are meant as information to the teacher about which of the capital letters and lower case
letters the students can recognize and link.
Test part 4:
To identify the initial sound: This measure the students' ability to do
exactly this. The teacher reads a word and asks the students to identify the rst sound
of the word and write this down. There is a picture supporting the word in the student
book. The students' results on this test should be scored by the teacher.
Test part 5:
synthesis.
To draw together sounds: Maps the students' abilities in phonological
Each part of this test contains a sequence with four pictures that illustrate
dierent words.
The teacher instructs the students by rstly presenting the word that
illustrate each picture and thereafter the target word, sound by sound, with a break
between every sound. The task of the students is to carry through the synthesis process
and determine which picture that goes with the target word.
The students' results on
this test should be scored by the teacher.
Understanding words
Test part 6:
To write words: Consists of a word dictation where each word is presented
for the students in a sentence. This test part comprises 8 sentences in total. The students'
results on this test should be scored by the teacher.
Test part 7:
To read words implies that the students should compare an illustration
with four written words and subsequently identify the word that ts with the illustration.
The students should identify as many word as possible (total possible words 19) within
34
ve minutes. The students' results on this test should be scored by the teacher.
Understanding sentences
Test part 8:
To read sentences consists of nine sub-parts. In each part the student
reads a sentence and mark the picture that illustrates the entire content of the sentence
among four alternative pictures with similar content. The length of the sentences increases
from two to ve words as the test proceeds. The students should link as many pictures
and sentences as possible within ve minutes. The students' results on this test should
be scored by the teacher.
A.3 Background information about the mathematics test
Every teacher responsible for carrying out tests in his/her class receives a teacher's
instruction manual. The following text about the test is from this manual of Education
[2011a].
This test consists of nine pages with several dierent tasks (se below).
The points
scored on each page should be added together. The critical threshold for the mathematics
test is based on the aggregated sum of points.
Page 1: Maps the students' ability to count, as well as if they know the numbers and
can link a number of items to a certain gure
Page 2: Maps if the students understand the idea equally many, i.e. that they can
compare the number in two dierent countable sets
Page 3: Investigates if the students' can rank the numbers in two dierent countable
sets and if they understand the concept most
35
Figure A2: Page 3
Page 4: Maps the students' knowledge about a sequence of numbers (linear)
Page 5: Maps the students' knowledge about a sequence of numbers (linear), if they can
count forward and backward from a given number, and if they understand the sequence
concepts prior to and subsequent to
Page 6: Investigates the students' knowledge of a series of numbers and the ranking
of given numbers within the series
Page 7:
Tests the students' ability to split a number into two other numbers (i.e.
4=1+. . . )
Page 8: Tests the students' ability to solve text assignments. The assignments have
dierent additive structures and they deal with both addition and subtraction (but rather
low numbers)
Page 9:
Maps the students' ability to add two numbers.
The additive structure
deal with the combination of two sums of money, to allow the students to employ their
36
knowledge about money and coins to arrive at the correct answer
37
Statistics Norway
Postal address:
PO Box 8131 Dept
NO-0033 Oslo
Office address:
Akersveien 26, Oslo
Oterveien 23, Kongsvinger
E-mail: [email protected]
Internet: www.ssb.no
Telephone: + 47 62 88 50 00
ISSN: 1892-753X
Design: Siri Boquist

Similar documents